//cc @nodejs/modules-active-members

The three failing tests in this PR are:

• ./node ./test/addons/async-hooks-promise/test.js the async hook promise injection is not being removed despite being disabled in https://github.com/nodejs/node/pull/29848/files#diff-e6db408e12db906ead6ddfac3de15a6fR314. @jasnell would value your input on this one if you have any suggestions here.
• ./node ./test/addons/make-callback-recurse/test.js, ./node ./test/node-api/test_make_callback_recurse/test.js both seem due to subtle timing changes with the switch to a promise-based bootstrap.

Adding WIP label based on PR description. Feel free to remove when appropriate, swap out with blocked label, or whatever. No strong opinions here. Just seems like the right thing to me. (The WIP label helps me in my workflow with the repository, so it's not just cosmetic for me, if that matters. But again, no strong opinions on this particular PR in that regard.)

On @joyeecheung's advice I've gone ahead and changed the way the async bootstrap works. Instead of always applying an async bootstrap, the async bootstrap and all promises associated will only apply lazily when the ESM loader is actually in use. This way loading CommonJS retains a fully sync bootstrap that is completely backwards compatible with no timings changes or task queue differences.

This also removes the dependence of this work on #29848.

All tests are now passing! Further review welcome.

I've split out the bootstrap changes into a separate PR at #29937, and based this unflagging to that PR for a simpler diff.

We likely want #29974 to land as well

@nodejs/modules anyone have an issue with renaming --loader to --experimental-loader before we land this? I can open a PR (and 12.x backport), but I'd be concerned with unflagging and not being explicit about this